Improving Recession Probability Forecasts in the U.S. Economy

Size: px
Start display at page:

Download "Improving Recession Probability Forecasts in the U.S. Economy"

Transcription

1 Improving Recession Probability Forecasts in the U.S. Economy Munechika Katayama Louisiana State University This Draft: April 26, 2009 First Draft: July 2008 Abstract There are two margins to improve forecasting models of the U.S. recession probability: including additional variables and using a different functional form. Using out-of-sample and cross validation methods, I systematically compare the performance of various forecasting models that differ in terms of variables included and functional forms used. I find substantial gains from including additional variables, such as the S&P 500 and non-farm employment growth, together with the Treasury term spread. In addition, there is a room to further improve forecasting accuracy by utilizing a non-normal cumulative distribution function. I also examine this possibility by using the generalized Edgeworth expansion, which enables us to explore a wider set of skewness and excess kurtosis. We can obtains further gains by allowing for more flexibility in the functional form. Keywords: Recession Probability; Forecasting; Generalized Edgeworth Expansion. JEL Classification: C25, C53, E32, E37. I am indebted to James Hamilton for his encouragement and helpful comments. For helpful comments and suggestions, I thank Hiroaki Kaido and participants at the Ninth Annual Missouri Economics Conference and the UC Riverside Conference on Business Cycles. This paper is based on Chapter 3 of my doctoral dissertation at UCSD. 1

2 1 Introduction There are many attempts to forecast recessions in the U.S. economy. Information contained in forward-looking variables can be used for making predictions about the future state of the economy. Stock and Watson (1989) report that the slope of the Treasury yield curve provides useful information in constructing their leading indicator. Estrella and Mishkin (1998) examine the usefulness of financial variables in making predictions about future recession probabilities at various forecasting horizons. They find that the slope of yield curve is the best single predictor of future recession probabilities. 1 Other studies also utilize the predictive content of the term spread in order to forecast recession probabilities. 2 Although developments in the literature achieve some success in making predicting recession, there is a room to improve forecasting accuracy. To obtain further improvements, there are two margins that we can exploit. One is the question of selecting a set of predictor variables. The other is a choice of a functional form. Typical studies on forecasting recession probabilities utilize a simple binary response model, such as a probit or logit model. In this paper, I will explore the importance of the two margins in improving forecasting accuracy. Within the class of probit/logit models, there is favorable evidence for benefits from including additional information, other than the term spread. For example, Estrella and Mishkin (1998) find some evidence on usefulness of including an additional variable, such as a New York Stock Exchange index or GDP growth. Wright (2006) reports that including the level of the federal funds rate together with the term spread results in better forecasting performance than the term spread alone. 3 King, Levin, and Perli (2007) recently find superior predictive power of the 5-year corporate credit spread on AA-rated firms compared with the 10-year-3-month Treasury term spread over the Great Moderation period (since the mid 1980 s). Furthermore, they also report that once the corporate spread is augmented by the term spread, the forecasting performance dramatically improves by reducing false positive predictions of recession. There are also some attempts to extend a simple probit model in the recent literature. For ex- 1 Their finding is consistent with a simple rule of thumb that the Treasury yield curve inversions or negative term spreads are likely followed by recessions in subsequent periods and the term spread has been cited as a leading economic indicator. 2 Other studies include Dotsey (1998), Birchenhall, Jessen, Osborn, and Simpson (1999), Estrella, Rodrigues, and Schich (2003), Stock and Watson (2003), Clements and Galvão (2006), Rudebusch and Williams (2008). 3 He also finds some evidence favorable for controlling for a term premium proxy. 2

3 ample, Chauvet and Potter (2005) extend a probit model in a way that it allows for business-cycle dependent coefficients (i.e., multiple breaks) and/or autocorrelated errors. Kauppi and Saikkonen (2008) develop dynamic probit models that include lagged explanatory variables and lagged recessionary dummies. Dueker (2005) presents a framework (a Qual VAR) that enables us to treat a qualitative variable as endogenous in a typical VAR framework, so that we can obtain dynamic forecasts of the qualitative variable. Going beyond the typical univariate probit model with the term spread, such as allowing additional features in a forecasting model and/or including additional predictor variables, most likely would improve in-sample forecasting performance. However, at the same time, we also have to worry about a potential over-fitting problem and we want to make a forecasting model as parsimonious as possible. Hansen (2008) examines the pitfalls of relying on in-sample fit. He shows that too good in-sample fit (over-fitting) tends to be associated poor out-of-sample fit and that model selection based on in-sample fit is not reliable. In addition, he demonstrates that tendencies of yielding spurious results are much higher if we use in-sample evaluation and are more pronounced when we compare performance of a large number of alternative models. For example, Sephton (2001) applies non-linear and non-parametric methods (multivariate adaptive regression splines) to forecasting recession probability. While in-sample fit with this very flexible approach shows great success, he finds that out-of-sample predictions are relatively not helpful. In this paper, for these reasons, I will use out-of-sample and cross validation methods to examine the importance of the two margins that can contribute to improve forecasting performance. First, I systematically compare the forecasting performance of 6-month-ahead recession probability predictions in the U.S. economy. I look at all possible combinations of 32 variables up to trivariate models with 6 different functional forms. Results highlight the importance of variable selection. Finding the best combination of predictor variables greatly improve forecasting performance. Especially, among variables considered, the combination of the term spread, changes in the S&P 500 stock price index, and growth rate of non-farm employment is found to achieve the best forecasting accuracy. Furthermore, additional gains can be obtained by using non-normal cumulative distribution functions. These additional features help improve forecasting accuracies by amplifying positive signals during recessions and also by dampening false positive alarms during expansions. However, there is some mixed evidence on what features of functional form are necessary 3

4 in order to improve forecasting performance. Using known CDFs amounts to imposing restrictions on the combination of skewness and excess kurtosis. Thus, I proceed to allow for more flexibility in the functional form by utilizing the generalized Edgeworth expansion of Jarrow and Rudd (1982), which allows us to explore a wider space of skewness and excess kurtosis. By using the genenelized Edgeworth expansion, we can obtain further improvement in out-of-sample forecasting accuracy. Findings in this paper complements other developments in the literature mentioned above. We can easily include additional variables into other extensions to probit models and also the more flexible functional form can be incorporated into a dynamic model or it can include multiple breaks, in principle. The rest of the paper is organized as follows. Section 2 lays out basic framework of the study. I present basic binary response models for forecasting recession probability, including the discussion of variables to be used and the method of evaluating forecasting performance. Section 3 presents empirical results and performs robustness check. Section 4 introduces the generalized Edgeworth expansion and applies it to making predictions of the U.S. recession probability. Finally, Section 5 concludes. 2 Basic Framework 2.1 Recession Probability Forecasting Model Let y t represent an NBER recession binary variable, which equals 1 when the economy is in recession in month t and equals 0 in expansion. 4 Typical models of forecasting h-period-ahead recession probabilities using the information available at time t assume that Prob(y t+h = 1 x t ) = F (β x t ), (1) where F ( ) is a monotonically increasing function, whose range is the unit interval, β is a vector of coefficients associated with a vector of predictors x t = [1, x 1,t,, x k,t ], and k is the number of 4 It is important to mention that, in reality, the NBER s announcement often involves a significant time lag. This is one drawback of relying on the NBER s decision. We can employ alternative definitions of business cycles, such as more than two consecutive quarters of negative output growth as a recession. Alternatively, we can utilize an algorithm proposed by Chauvet and Hamilton (2006), which is motivated by the delay in the NBER s announcement, for nowcasting, in order to determine near-real-time values of y t. However, since it is the NBER s decisions that is most widely used in the academia as well as the public, I will use it to define the state of the U.S. economy. 4

5 variables included. It is commonly assumed that y t+h is a conditionally independent Bernoulli random variable, so that the likelihood function is given by: T [ L = F (β x t ) ] y t+h [ 1 F (β x t ) ] 1 y t+h. (2) t=1 In an empirical analysis, I will set h = 6 and focus on 6-month-ahead predictions. In this formulation, predicting recession probability involves two issues that are possibly related to each other. The first one is to choose a set of predictor variables, so that we can obtain useful information from data. A workhorse predictor in the literature is the Treasury term spread between 10-year and 3-month bonds, which is due to the finding in Estrella and Mishkin (1998). Given a choice of F ( ), finding a better combination of predictor variables obviously helps improve forecasting accuracy if those variables contain different information and jointly provide useful signals. Since there are no a priori variable selection procedures available, our approach is to try all possible combinations of variables in order to find a better combination of variables that are helpful and stable in forecasting recession probabilities. Although knowing the best single predictor is helpful, it is unlikely that we can obtain some insights about a better combination of variables by just looking at forecasting performance of single predictors because those variables that show relatively good forecasting performance tend to contain similar information. The second issue is how to translate signals into a probability measure between 0 and 1, which is related to a shape of F ( ). In order to guarantee that F ( ) is monotonically increasing and takes values between 0 and 1, we typically use a known cumulative distribution function (CDF). A popular choice is to use either the Standard Normal CDF (a probit model) or the Logistic CDF (a logit model), or some extensions to those (e.g., Chauvet and Potter, 2005; Kauppi and Saikkonen, 2008). However, the shape of F ( ) is not necessarily restricted to typical ones. In principle, the CDF of any continuous probability random variable will suffice. Differences in the shape of CDFs can be characterized in terms of skewness and excess kurtosis. It should be noted that it is not appropriate to use terms skewness and excess kurtosis here because we are not talking about characteristics of underlying statistical distributions, but just utilizing functional forms. However, for expositional simplicity, I will use those terms in describing 5

6 F (x) 1 F (x) x 0 x (a) Excess Kurtosis > 0 and Skewness = 0 (b) Excess Kurtosis > 0 and Skewness > 0 Figure 1: Role of Skewness and Excess Kurtosis the shape of a CDF. The consequence of allowing skewness and excess kurtosis is illustrated in Figure 1. Higher excess kurtosis makes a CDF steeper around the median of β x, as depicted in the left panel of Figure 1. On the other hand, allowing non-zero skewness makes a CDF asymmetric around F (β x) = 0.5, as illustrated in the right panel of Figure 1. With zero skewness, F (x) = 1 F ( x) for any x R. However, non-zero skewness implies F (x) 1 F ( x). 5 In order to understand what features of a CDF are helpful in improving forecasting accuracy, I will consider 6 different CDFs. Table 1 summarizes characteristics of the CDFs considered. In addition to the Standard Normal CDF and the Logistic CDF, I consider Student-t, Laplace, Gumbel, and Type III Generalized Extreme Value (GEV3). All location parameters and scale parameters, if applicable, are set to 0 and 1, respectively. These CDFs are chosen and configured to incorporate non-zero skewness and/or higher excess kurtosis that the Standard Normal CDF does not have. The first four CDFs have zero skewness and differ in terms of degree of excess kurtosis. The Logistic CDF has excess kurtosis of 1.2. The degrees of freedom parameter for Student-t is set to be 6.5, such that its excess kurtosis equals 2.4. The excess kurtosis of Laplace is 3. The last two CDFs also have positive skewness. For the Gumbel CDF, skewness is equal to and excess kurtosis is 2.4. The shape parameter of the Type III GEV, s, is set such that its excess kurtosis equals 3 (i.e., s = ). The associated skewness becomes Thus, for a given set of predictor variables, I will be able to infer the importance of allowing for excess kurtosis 5 In general, changing location and scale parameters does not affect forecasting results and they are fixed to avoid an identification problem. Changing the location parameter just changes the estimate of the constant term and changing the variance just results in different scaling of coefficients β. 6

7 Table 1: List of the Cumulative Distribution Functions Considered Type of CDF F (x) Skewness Excess Kurtosis 1 x ( ) u 2 Standard Normal exp du 0 0 2π 2 Logistic exp(x) 1 + exp(x) Student-t with ν = 6.5 Laplace x Γ( ν+1 2 ) νπγ( ν 2 ) ) (1 (ν+1) + u2 2 du ν 1 [1 + sgn(x) {1 exp ( x )}] Gumbel 1 exp { exp( x)} Type III GEV with s = { exp (1 + sx) 1/s} Note: All location parameters are set to be 0 and all scale parameters are set to be 1. ν represents degrees of freedom parameter for the Student-t distribution. Γ( ) is a gamma function. Excess kurtosis of Student-t is given by 6/(ν 4) for ν > 4. Skewness of the Gumbel distribution is 12 6ζ(3), where π 3 ζ( ) is a zeta function. s denotes a shape parameter of the generalized extreme value distribution. For the Type III GEV, skewness is given by Γ(1 3s)+3Γ(1 s)γ(1 2s) 2(Γ(1 s))3 (Γ(1 2s) (Γ(1 s)) 2 ) 3/2 Γ(1 4s) 4Γ(1 s)γ(1 3s)+6Γ(1 2s)(Γ(1 s)) 2 3(Γ(1 s)) 4 (Γ(1 2s) (Γ(1 s)) 2 ) 2. and excess kurtosis is give by in F ( ) by looking at results based on the first four CDFs. Furthermore, comparing the Gumbel with the Student-t or the Type III GEV with the Laplace enables us to obtain some insight on whether allowing positive skewness is helpful in improving forecasting performance Data Table 2 lists all 32 monthly variables considered in this paper. The sample period starts from January 1960 and ends at December It is chosen to maximize data availability and to include as many recession episodes as possible. As a result, the sample covers 8 post-war recessions. This is important in choosing a set of predictor variables that are robust. It is not so difficult to find a particular variable combination that can explain previous recessions ex post. However, there is no guarantee that a good predictor for a particular recession works well in predicting another one. The data set contains the term spreads, the credit spreads, various interest rates, employment 6 It is also possible to introduce negative skewness by using the Type III GEV. However, preliminary estimation indicated that forecasting performance based on the Type III GEV with negative skewness is inferior to others in general. So, I do not report their results. 7

8 Table 2: List of Variables Predictor Description Info. Lag Interest Rates 0 FF Federal Funds rate 0 3M 3-month Treasury Bill rate 0 5Y 5-year Treasury Bond rate 0 10Y 10-year Treasury Bond rate 0 AAA Moody s corporate bond yield, AAA 20 years or longer 0 AA Moody s corporate bond yield, AA 20 years or longer 0 A Moody s corporate bond yield, A 20 years or longer 0 Term Spreads TS10YFF 10Y-FF Treasury term spread 0 TS10Y3M 10Y-3M Treasury term spread 0 TS10Y5Y 10Y-5Y Treasury term spread 0 Credit Spreads CSAAA AAA - 10Y spread 0 CSAA AA - 10Y spread 0 CSA A - 10Y spread 0 Employment Data EMP Non-agricultural employment (log-differenced) 0 CEMP Civilian employment (log-differenced) 0 UICLAIM Initial unemployment insurance claims (log-differenced) 1 UNEMP Unemployment rate 0 UNEMPD Changes in unemployment rate 0 HOURS Average weekly hours in manufacturing (log-differenced) 0 Stock Price Indices DJ30 Dow Jones 30 average (% changes over 3 months) 0 SP500 S&P 500 stock price index (% changes over 3 months) 0 Monetary Aggregates M0 Monetary base (log-differenced) 1 M1 M1 (log-differenced) 1 M2 M2 (log-differenced) 2 Other Macroeconomic Variables CLI11 Composite leading indicators (11 series, 1987=100, log-differenced) 1 CPI CPI, all urban, all items (log-differenced) 1 EXP Consumer expectation ( = 100) 0 EXPD Changes in consumer expectation 0 HOUSE New private housing units authorized by building permits (log-differenced) 1 VENDOR Vendor performance (slower deliveries diffusion index, %) 0 INCOME Personal income less transfer payments (log-differenced) 2 IP Industrial production (log-differenced) 1 SALES Manufacturing & trade sales (log-differenced) 1 Note: Information lag is measured at the end of month. Strictly speaking, those employment data with zero information lag and vendor performance are not available at the end of the month. However, they will be available at the very beginning of the next month and there are virtually no considerable lags. Thus, they are categorized in the zero information lag variable. 8

9 data, stock price indices, monetary aggregates, and other macroeconomic variables. Most of those are investigated in the earlier studies or used in constructing a composite leading indicator. It also includes variables, to which the NBER Business Cycle Dating Committee pays particular attention in deciding business cycle peaks and troughs. Since Kane (2008) documents the usefulness of employment data in predicting occurrence of recessions, the data set includes employment-related variables as well. Recently, King, Levin, and Perli (2007) report that, in the period of the Great Moderation, the credit spread on AA-rated firms has particularly good forecasting performance. However, the corporate bond yields used in their study (maturity of 5-year and 10-year) do not cover the entire sample. Since there are not many recessionary periods, especially after the 1980 s, using shorter sample periods may have considerable effects on evaluating forecasting performance. Thus, I have decided not to include the credit spreads that King, Levin, and Perli (2007) have used. Instead, as a crude proxy, the data set include spreads between Moody s AAA-, AA-, or A- rated corporate bond yield (20 years or longer) and the 10-year Treasury bond yield. 7 When we focus on financial variables as predictors, we do not need to consider a gap between when an observation is made and when it is available for forecasting. However, as shown in Table 2, some series are not reported immediately and we need to take account of the information lag, in order to accurately assess forecasting models. In this paper, any variable z t represents the latest data on z available at month t, instead of an observation at month t. For example, industrial production (IP) has 1 month of the information lag. So, IP 2000:01 refers to the industrial production data on December Because the total number of models increases exponentially as the number of variables included (k) increases, I will restrict my attention to k 3. This will result in examining a total of 3 k=1 32! k!(32 k!) 6 = forecasting models. 2.3 Evaluating Forecasting Performance In order to evaluate various forecasting models, I will primarily focus on out-of-sample results. Hansen (2008) shows that in-sample and out-of-sample fits are negatively correlated, which implies 7 Since the 20-year Treasury bond yield data has discontinuity between January 1987 and September 1993, I use the 10-year T-bond rate, instead. Thus, precisely speaking, this credit spread is a combination of the true credit spread and the term spread. 9

10 that good in-sample performance is not a useful indicator of out-of-sample accuracy and that relying on in-sample fit is highly misleading. This over-fitting problem is particularly important, since the likelihood of obtaining spurious results is more pronounced when we search a large number of alternative models. For this reason, I will focus on recursive out-of-sample forecasting evaluation and I will use cross validation as a robustness check. For recursive (pseudo) out-of-sample forecasting exercises, the out-of-sample prediction starts from January 1989 and ends at the end of the full sample. The out-of-sample period covers the last three recessions in the U.S. economy. By adding observations one by one, I estimate a forecasting model again to produce a forecast for the next month. In reality, when we forecast future recession probability, we are not certain about the true state of the economy in recent months. It is because the decision of the NBER Business Cycle Dating Committee typically involves substantial time lag. To be realistic, I assume that forecasters do not know the true state of the economy for a year and assume that y t = y t 1 = = y t 11 = y t 12. In other words, a forecaster assumes that the economy is in the same state as a year ago. Following Clements and Galvão (2006), I will use three different measures for evaluating accuracy of out-of-sample recession probability predictions. The first measure is the probabilityanalogue of mean squared error, the quadratic probability score (QPS), which is commonly used in evaluating probability forecasts. The QPS is defined as QP S = 2 T T (ˆp t y t ) 2, (3) t=1 where ˆp t is the recession probability forecast month t. The QPS takes values between 0 and 2 and smaller value indicates more accurate forecasts. The second measure is the log probability score (LPS), which is defined as LP S = 1 T T { yt log( ˆp t ) + (1 y t ) log(1 ˆp t ) }. (4) t=1 The LPS ranges from 0 to + and a smaller value corresponds to more accurate predictions and penalizes larger mistakes more heavily than the QPS. 10

11 The last measure is the Kuipers Score (KS), which is given by KS = T t=1 y t1 [(ˆpt>0.5)] T t=1 y t T t=1 (1 y t)1 [(ˆpt>0.5)] T t=1 (1 y, (5) t) = hit rate false rate, (6) where 1 [ ] is an indicator function that equals 1 if its argument is true and 0 otherwise. The KS calculates the difference between the hit rate and the rate of false signals by using 50% probability of recession as a cutoff. The KS takes values between 1 and 1. A score of 1 corresponds to making perfect predictions. The KS evaluates the recession predictions from a slightly different aspect, compared with other two measures. Even when recession predictions never show strong indications (say, higher than 50% probability), it is possible to have seemingly good results based on the QPS and LPS. The KS discounts such weak predictions. In this sense, the KS captures the strength and accuracy of predictions by using the 50% probability cutoff. There is a potential problem of just relying on the recursive out-of-sample exercises described above, especially in the context of recession probability forecasting. Since there are not many recession episodes in the out-of-sample period, it is possible to select a forecasting model that has particularly good performance for the last three recessions, but not a next recession. In order to robustify the results, I will also carry out a cross-validation type exercise, called leaving 2-years out. The detailed procedures of the leaving 2-years out exercises are as follows. Let S = {(y t, x t h ) : t = 1,, T } denote a full sample and L τ = {(y t, x t h ) : t = τ 12,, τ + 12} represent a set of excluding observations. For each τ = 13,, T 12, (i) Take E τ = S \ L τ as a training sample. (ii) Estimate parameter values β τ based on E τ. (iii) Make a prediction for y τ by using β τ and x τ h and store it. (iv) Repeat steps (i) (iii). Then calculate the QPS, LPS, and KS based on {ŷ τ } T 12 τ=13. It is necessary to find a forecasting model whose forecasting performance is stable over time and is robust to different recessions. The recursive out-of-sample forecasting exercises seem to be 11

12 a more realistic setup. However, at the same time, it could be vulnerable to serially correlated errors. Given the conditional independence assumption, the leaving 2-years out exercise provides a completely valid evaluation and hopefully picks up a forecasting model whose performance is robust across different recession episodes. For this reason, it is very important to draw conclusions based on two different ways of evaluating forecasting performance. 3 Results 3.1 Univariate Probit Models First, we will start off by looking at forecasting performance within a class of univariate probit models of predicting 6-month-ahead recession probabilities (h = 6) as a benchmark. Table 3 summarizes the out-of-sample forecasting performance of univariate probit models. Overall, the out-of-sample forecasting performance evaluated by the QPS and LPS shows broadly similar patters for the rankings of predictor variables. While the QPS selects the composite leading indicator (CLI11) as the best predictor within univariate probit models, the LPS chooses the term spread between 10-year Treasury bond yield and the Federal Funds rate (TS10YFF) as the best predictor. Based on both the QPS and LPS, 3-month change in S&P 500 (SP500) appears to be one of best predictors. It is important to point out that these variables outperform the widely used recession probability predictor, the term spread between 10-year and 3-month Treasury yields (TS10Y3M). Given the forecasting horizon of 6 months, superiority of TS10YFF over the conventionally used TS10Y3M is true not only in the out-of-sample forecasts, but also in the entire sample (detailed in-sample results are not reported here). In contrast to better performance of these term spreads the credit spreads (CSA, CSAA, and CSAAA) have poor forecasting performance. Among other variables, CPI inflation (CPI) and University of Michigan s index of consumer expectation (EXP) have relatively good performance based on both the QPS and LPS. Some of employment related variables, such as changes in unemployment rate (UNEMPD) and growth rate of the initial unemployment insurance claims (UICLAIM), might contain useful information. However, other employment related variables, those variables that the NBER Business Cycle Dating Committee is paying attention to, and monetary aggregates have considerably poorer results. 12

13 Table 3: Variable Rankings with Univariate Probit Models QPS Ranking LPS Ranking KS Ranking CLI TS10YFF TS10YFF TS10YFF CLI TS10Y3M SP SP UNEMPD TS10Y3M TS10Y3M EXPD CPI EXP EMP EXP CPI IP UNEMPD UICLAIM SALES M UNEMPD UNEMP FF EXPD CEMP EXPD EMP HOURS UICLAIM IP M IP SALES A EMP UNEMP FF UNEMP CEMP HOUSE A HOURS AA SALES M AAA CEMP A VENDOR AA FF M AAA HOUSE M HOURS AA Y M AAA Y HOUSE VENDOR CSA Y M CSAA Y M CSAAA VENDOR INCOME M CSAAA Y CLI CSAA Y EXP CSA CSA CPI M CSAA UICLAIM INCOME M INCOME TS10Y5Y CSAAA TS10Y5Y M TS10Y5Y SP Although the rankings of out-of-sample forecasting accuracy based on the QPS and LPS are broadly consistent each other, some discrepancy results from the fact that the LPS penalizes larger mistakes more heavily. For instance, the levels of short-term interest rate, such as the Federal Funds rate (FF) and the 3-month Treasury bill yield (3M) are somewhat informative based on the QPS, the LPS suggests that they are relatively poor predictors. Figure 2 compares 6-month-ahead out-of-sample predictions that are selected as the best based on the QPS and LPS. The shaded areas indicate the NBER recession periods. Visual inspection of these picture suggests a potential pitfalls in relying on one particular measure of forecasting accuracy. Although predictions from CLI11 have the smallest errors in terms of the quadratic loss, 13

14 1 Univariate Probit with CLI Univariate Probit with TS10YFF Figure 2: Out-of-Sample Predictions from Univariate Probit Note: The shaded areas represent the NBER recessions. Horizontal axes measures recession probabilities. they are not useful forecasts at all in practice, since they appear to be just noisy signals and they do not show clear contrasts between recessions and expansions. On the other hand, out-of-sample predictions with TS10YFF have better distinctions about the state of the economy. Although overall performance is better than others, there are a couple of issues with predictions with TS10YFF. For the recession, the predictions made by the term spreads miss the timing of the recession and they do not have strong signals. This difficulty is also noted in other studies. See Stock and Watson (2003) for more detailed discussions on this issue. Furthermore, the magnitude of the recession signal is not strong enough, so that it is difficult to distinguish between a true signal and a false alarm. Peaks in the recession probability predictions never exceed 50%. For example, it 14

15 is extremely difficult in real time to distinguish these two humps during the 1990 s from those for the true signal for the 2001 recession. Overall pattern of this out-of-sample predictions is promising. However, it would be better if predictions exhibit stronger signals during true recessions. This gives us an incentive to pay attention to another measure of forecasting accuracy, the KS, which takes account of the accuracy and strength of predictions. The out-of-sample forecasting performance evaluated by using the KS gives us a completely different picture. Most predictor variables have zeros for the KS. In other words, they usually do not give us strong predictions about the occurrence of future recessions or correct predictions are largely offset by false predictions. In the worst case, we will get more false signals, which indicate more than 50% probability of a recession during an expansion, than correct ones. Such a situation happens to many of predictors that are considered to be good based on the QPS and LPS, such as CLI11, SP500, EXP, and CPI. 3.2 Other Univariate Models Now we will look at the performance of univariate models based on alternative CDFs, which allow positive skewness and/or excess kurtosis. Table 4 summarizes the ranking of univariate models based on out-of-sample fit. 8 As in the probit models, the QPS and LPS suggest that those models with CLI11, TS10YFF, SP500, and TS10Y3M have relatively good forecasting performance. For CLI11, while adding skewness marginally improves forecasting accuracy in the out-of-sample forecasting exercises compared to the probit model, allowing higher excess kurtosis alone deteriorates performance. Especially, the CLI11 with Laplace, which has the highest excess kurtosis with zero skewness is the worst among the models with CLI11. In fact, it is worse than the second-best predictor in probit models, TS10TFF. We can observe a similar pattern in those models with SP500. Within the class of univariate models with a term spread measure, Normal works better than alternative CDFs. In fact, according to the LPS, Normal with TS10YFF has the best result. There are some role played by using non-normal CDFs in improving forecasting accuracy. It is not easy to generalize the role of different functional forms and the benefit of allowing skewness 8 Due to space limitation, hereafter top 20 forecasting models are shown. All results are available from the author upon request. 15

16 Table 4: Top 20 Univariate Models QPS Ranking LPS Ranking KS Ranking CDF x 1 QPS CDF x 1 LPS CDF x 1 KS Gumbel CLI Normal TS10YFF Laplace TS10Y3M GEV3 CLI GEV3 CLI Laplace CLI Normal CLI Gumbel CLI Laplace EXP Logistic CLI Normal CLI Gumbel SP Student-t CLI GEV3 TS10YFF GEV3 SP Normal TS10YFF Logistic TS10YFF Laplace SP GEV3 TS10YFF Student-t CLI Normal TS10YFF Logistic TS10YFF Logistic CLI GEV3 TS10YFF Gumbel TS10YFF Student-t TS10YFF Logistic TS10YFF Student-t TS10YFF Laplace TS10YFF Gumbel TS10YFF Gumbel SP Laplace CLI Student-t TS10YFF Laplace CLI Gumbel TS10YFF Laplace TS10YFF Laplace TS10YFF Gumbel SP Normal TS10Y3M GEV3 SP GEV3 SP GEV3 TS10Y3M Normal SP Normal SP Gumbel TS10Y3M Student-t SP Logistic SP Logistic TS10Y3M Logistic SP Student-t SP Student-t TS10Y3M Laplace SP Laplace SP GEV3 EXP Normal TS10Y3M Gumbel TS10Y3M Gumbel EXP GEV3 TS10Y3M Normal TS10Y3M Laplace UNEMPD and/or excess kurtosis may be variables specific. However, we can see clear effects of using non- Normal CDFs if we look at the ranking based on the KS. Within the class of univariate models, most of models scores non-positive values for the KS. In other words, most of predictors do not make correct and strong predictions, in terms of the 50% probability cutoff. However, by using non-normal CDF we can improve the KS. There are 6 models with positive values for the KS. Especially, using Laplace is likely to contribute to producing stronger and more accurate signals, net of false alarms. Especially, CLI11, EXP and SP500 all have negative values for the KS in the probit models. Using non-normal CDF can alter bad performance into better predictions. 3.3 Bivariate Models Now we turn out attention to forecasting results based on bivariate models in order to see the importance of additional information. Table 5 shows the top 20 bivariate forecasting models based on the QPS, LPS, and KS. The combination of TS10YFF and SP500 is the best predictor variables based on the QPS and LPS. Both the QPS an LPS suggest that it is important to include a measure of term spread (either TS10YFF or TS10Y3M) and the relative importance of CLI11 decreases. Interestingly, there are 16

17 Table 5: Top 20 Bivariate Models QPS Ranking LPS Ranking KS Ranking CDF x1 x2 QPS CDF x1 x2 LPS CDF x1 x2 KS Gumbel TS10YFF SP Normal TS10YFF SP Logistic EXP UNEMP GEV3 TS10YFF SP Logistic TS10YFF SP Student-t EXP UNEMP Normal TS10YFF SP Student-t TS10YFF SP Laplace EXP UNEMP Student-t TS10YFF SP GEV3 TS10YFF SP Gumbel TS10Y3M SP Logistic TS10YFF SP Laplace TS10YFF SP GEV3 TS10Y3M SP Laplace TS10YFF SP GEV3 TS10YFF EMP Normal TS10Y3M SP Gumbel TS10Y3M SP Gumbel TS10YFF SP Logistic TS10Y3M SP GEV3 TS10Y3M SP Normal TS10YFF EMP Student-t TS10Y3M SP Normal TS10Y3M SP Logistic TS10YFF EMP Laplace TS10Y3M SP Logistic TS10Y3M SP Student-t TS10YFF EMP Normal TS10Y3M EMP Student-t TS10Y3M SP Gumbel TS10Y3M EXP GEV3 TS10Y3M EMP Laplace TS10Y3M SP Gumbel TS10YFF EMP Gumbel TS10Y3M EMP Gumbel CLI11 SP Gumbel TS10Y3M SP Laplace UICLAIM SP GEV3 CLI11 SP GEV3 TS10Y3M SP Gumbel EMP SP Laplace EXP UNEMP Normal TS10Y3M SP GEV3 EMP SP Normal TS10YFF EMP Laplace TS10YFF EMP Normal EMP SP GEV3 TS10YFF EMP GEV3 TS10Y3M EXP Student-t EMP SP Normal CLI11 SP Logistic TS10Y3M SP Logistic EMP SP Logistic TS10YFF EMP Student-t TS10Y3M SP Logistic CLI11 SALES Student-t TS10YFF EMP Gumbel CLI11 SP Student-t CLI11 SALES

18 some variables that have relatively poor performance in univariate models and produce good outof-sample forecasting accuracy, together with the term spread measure. According to the rankings based on the univariate models, SP500 and EMP are not a useful single predictor. Especially, EMP has relatively poor out-of-sample forecasting performance. However, all three measures of forecasting accuracy suggest that they are an important companion variable to other variables. In fact, that the best bivariate model is not a combination of the two best single predictors (CLI11 and TS10YFF). This suggests that the univariate ranking is not a helpful guide for choosing multiple predictors. The importance of SP500 together with a term spread measure is consistent with the finding of Estrella and Mishkin (1998). In our bivariate models, even without being combined with the term spread, some bivariate models that contain SP500 perform relatively well. However, it should be mentioned that King, Levin, and Perli (2007) do not find superiority of SP500 in conjunction with the term spread. Rather, they report that a combination of variables, which have better performance in univariate models, also perform better in bivariate models. 9 Based on the QPS, allowing positive skewness and excess kurtosis (Gumbel and GEV3) outperform the probit counterparts. However, introducing only excess kurtosis (Logistic, Student-t, and Laplace) worsens the QPS, compared with the probit. However, for the best bivariate models with TS10YFF and SP500, utilizing non-normal CDFs deteriorates forecasting accuracy, based on the LPS. In other words, it tends to some improvements in the sense of the quadratic loss, but at the same time, it is likely to produce larger mistakes. Especially, positive skewness makes the deterioration worse. Although forecasting performance generally improves by adding one more predictor, probably the biggest gain appears in the KS. In univariate models, only 6 models out of 192 have positive scores. In bivariate models, 294 models out of 3168 have at least positive values for the KS. Furthermore, the magnitude of the KS improves significantly. This means that by incorporating additional variables, we are likely to produce more accurate and stronger signals. Although the variable combination of TS10YFF and SP500 is the best based on the QPS and LPS, it does 9 This could be because of a couple of reasons. First, it could be attributed to the difference in periods used for the out-of-sample forecasting exercises and forecasting horizon. Second, it might be because of the fact that my data set does not include the credit spread measures that they use and perform quite well in their univariate models. Finally, it could be due to the difference in evaluating out-of-sample forecasting performance. They look at average out-of-sample predictions over two test periods, the 2001 recession and the post-2001 expansion. 18

19 not perform well in terms of the KS (KS = for Normal and KS = for Gumbel). Instead of using TS10YFF, the term spread with 3-month Treasury bill yields, together with SP500, gives us better results for the KS and provides consistently good performance across different forecasting accuracy measures. Relatively consistent performance can be also observed in the variable combination of TS10Y3M and EMP as well. 3.4 Trivariate Models and Overall Rankings Now we move on to the overall forecasting performance, including all of univariate, bivariate, and trivariate models with 6 different CDFs. Tables 6 lists the top 20 forecasting models out of forecasting models, based on different criteria and Figure 3 summarizes and compares overall performance of selected models. According to the out-of-sample forecasting performance, based on the QPS and the LPS, the Laplace with TS10YFF, SP500, and EMP is the best forecasting model. Regardless of the functional form used, this variable combination records outstanding out-of-sample performance. This results is consistent with the fact that in bivariate models, SP500 and EMP are found to be good companion variables to TS10YFF. Other models with TS10YFF, SP500, and VENDOR (vendor performance index), and those with TS10Y3M, SP500, and EMP also show relatively good forecasting performance based on the QPS and LPS. For the variable combination of TS10YFF, SP500, and EMP, there are some roles played by positive skewness and/or excess kurtosis in improving the forecasting accuracy. In terms of the QPS, allowing higher excess kurtosis without positive skewness is likely to outperform the probit counterpart. With the LPS, it is hard to see some clear patterns but there is non-trivial difference between the Laplace model and the probit model. However, it the term spread measure is replaced with TS10Y3M, the role of non-normal CDFs is completely flipped. For these models with TS10Y3M, SP500, and EMP, allowing additional features in the functional form deteriorates forecasting accuracy. In other words, the differences in the measures of the term spread (and interactions with other variables) affects the usefulness of the different functional form. Although there is favorable evidence that allowing excess kurtosis and skewness help improve forecasting accuracy, such a possibility can be variable specific. Out-of-sample forecasting accuracy evaluated by the KS suggests different ranking of models. 19

20 Table 6: Top 20 Models from Out-of-sample Forecasting Exercises QPS Ranking LPS Ranking CDF x 1 x 2 x 3 QPS CDF x 1 x 2 x 3 LPS Laplace TS10YFF SP500 EMP Laplace TS10YFF SP500 EMP Student-t TS10YFF SP500 EMP GEV3 TS10YFF SP500 EMP Logistic TS10YFF SP500 EMP Logistic TS10YFF SP500 EMP GEV3 TS10YFF SP500 EMP Normal TS10YFF SP500 EMP Normal TS10YFF SP500 EMP Student-t TS10YFF SP500 EMP Gumbel TS10YFF SP500 EMP Gumbel TS10Y3M SP500 EXP Gumbel TS10YFF SP500 VENDOR Gumbel TS10YFF SP500 EMP GEV3 TS10YFF SP500 VENDOR GEV3 TS10Y3M SP500 EXP Student-t TS10YFF SP500 VENDOR GEV3 TS10Y3M SP500 EMP Logistic TS10YFF SP500 VENDOR Normal TS10Y3M SP500 EMP Gumbel TS10YFF SP500 CEMP Gumbel TS10Y3M SP500 EMP Normal TS10Y3M SP500 EMP Normal TS10Y3M SP500 EXP Laplace TS10YFF SP500 VENDOR Logistic TS10Y3M SP500 EMP Gumbel TS10YFF SP500 TS10Y3M Student-t TS10Y3M SP500 EMP Gumbel FF SP500 5Y Gumbel TS10YFF SP500 EXP Normal TS10YFF SP500 VENDOR Student-t TS10YFF SP500 VENDOR GEV3 TS10YFF SP500 CEMP Logistic TS10YFF SP500 VENDOR Gumbel TS10YFF SP500 CLI GEV3 TS10YFF SP500 CEMP Gumbel TS10YFF SP500 UNEMPD Logistic TS10Y3M SP500 EXP Gumbel TS10YFF SP500 EXP Student-t TS10YFF SP500 CEMP KS Ranking CDF x 1 x 2 x 3 KS Gumbel TS10Y3M SP500 EMP GEV3 TS10Y3M SP500 EMP Logistic TS10Y3M SP500 EMP Student-t TS10Y3M SP500 EMP Laplace UNEMP EXP EMP Logistic UNEMP SP500 EMP Student-t UNEMP SP500 EMP Normal TS10Y3M SP500 EMP Laplace UNEMP SP500 EMP Gumbel UNEMP SP500 EMP GEV3 UNEMP SP500 EMP Laplace UNEMP EXP M GEV3 UNEMP EXP EMP Normal UNEMP EXP EMP Laplace TS10Y3M SP500 EMP Laplace UNEMP EXP UNEMPD Laplace UNEMP EXP IP Student-t UNEMP EXP EMP Logistic UNEMP EXP EMP Laplace UNEMP SP500 EXP

21 TS10YFF + SP500 + EMP TS10YFF + SP500 + VENDOR TS10Y3M + SP500 + EMP Others Gumbel 0.31 LPS GEV3 Laplace Student t Logistic Normal Logistic Normal GEV3 Student t Gumbel Laplace Student t Logistic Laplace Gumbel Normal GEV QPS Figure 3: Overall Performance of Selected Models in Out-of-sample Forecasting Exercises Note: Horizontal axis takes QPS and vertical axis measures the LPS. For both measures, the smaller the values are, the better the forecasting performance is. Thus, a southwest corner of the plot corresponds to the best forecasting model based on the QPS and the LPS. Top 20 models based on the QPS and LPS are plotted. 21

22 Table 7: Importance of the Two Margins Models QPS LPS KS Normal Laplace GEV3 Normal Laplace GEV3 Normal Laplace GEV3 Univariate * Bivariate * Trivariate * * Note: The univariate model includes TS10YFF. The bivariate model refers to models with TS10YFF and SP500 and the trivariate models add EMP to the bivariate models. Asterisks indicate the best model in each of univariate, bivariate, and trivariate models, based on different criteria. The best models according to the KS utilize information contained in TS10Y3M, SP500, and EMP. Especially, the one with Gumbel achieves the best result. For this combination of variables, positive skewness and excess kurtosis appear to be very helpful to improve the out-of-sample forecasting accuracy in terms of the KS. Good forecasting performance based on the QPS and LPS usually is not supported by the KS performance and vice versa. Except for the combination of TS10Y3M, SP500, and EMP, the variable combinations that are ranked relatively high in the KS ranking do not have good out-of-sample forecasting performance in terms of the QPS or LPS. In other words, making strong predictions during recessions involves risks of making positive false signals during expansions, which worsens the QPS and LPS. In this sense, the variable combination TS10Y3M, SP500, and EMP might be preferred since it has relatively stable performance across different measures of forecasting accuracy. 3.5 Discussion Forecasting performance based on all models indicates that the term spread measure (TS10YFF) is one of the most important variables, as widely documented in other studies. However, there are two margins to improve forecasting performance. One is to include additional predictors. By combining different information, we can increase accuracy of recession probability predictions. As demonstrated, a particular combination of variables greatly outperforms a typical univariate probit model with TS10YFF. Bivariate results suggest the importance of the stock price index (SP500) as a companion variable to the term spread measure. In addition, among the all combinations of variables considered in this paper, augmenting the term spread measure by SP500 and EMP results in the best out-of-sample forecasting performance. Another margin is to change a functional form of F ( ), in order to improve how to translate signals to a recession probability measure. 22

23 1 0.9 Univariate Probit Trivariate Probit Difference between Trivariate Laplace and Trivariate Probit Figure 4: Comparison of Out-of-Sample Predictions Note: Shaded areas indicate the NBER recession episodes. The top panel compares predictions from the univariate probit model with TS10YFF with those from the trivariate model with TS10YFF, SP500, and EMP. The bottom panel shows differences in predictions by changing the functional form. The bottom panel plots the difference between the trivariate Laplace and the trivariate probit. It is important to understand how these two margins contribute to better forecasting performance. Table 7 summarizes how forecasting performance improves by changing the two margins and illustrates different effects of the margins. We compare probit models with the Laplace and GEV3 counterparts. Provided that a forecaster knows which one is the best companion variable, including additional predictors has greater impact on improving all three measures of forecasting accuracy. However, it is possible to obtain non-negligible gains from changing the functional form. Although up to the bivariate models, probit specification works better than any other CDFs, changing the functional form has have bigger impacts, once EMP is included in addition to TS10YFF and SP

A Markov switching regime model of the South African business cycle

A Markov switching regime model of the South African business cycle A Markov switching regime model of the South African business cycle Elna Moolman Abstract Linear models are incapable of capturing business cycle asymmetries. This has recently spurred interest in non-linear

More information

Predicting Turning Points in the South African Economy

Predicting Turning Points in the South African Economy 289 Predicting Turning Points in the South African Economy Elna Moolman Department of Economics, University of Pretoria ABSTRACT Despite the existence of macroeconomic models and complex business cycle

More information

Forecasting U.S. Recessions with Macro Factors

Forecasting U.S. Recessions with Macro Factors Forecasting U.S. Recessions with Macro Factors Sebastian Fossati University of Alberta This version: May 19, 2015 Abstract Dynamic factors estimated from panels of macroeconomic indicators are used to

More information

B usiness recessions, as a major source of

B usiness recessions, as a major source of Regime-Dependent Recession Forecasts and the 2 Recession Michael J. Dueker B usiness recessions, as a major source of nondiversifiable risk, impose high costs on society. Since firms cannot obtain recession

More information

Predicting Bear and Bull Stock Markets with Dynamic Binary Time Series Models

Predicting Bear and Bull Stock Markets with Dynamic Binary Time Series Models ömmföäflsäafaäsflassflassflas ffffffffffffffffffffffffffffffffff Discussion Papers Predicting Bear and Bull Stock Markets with Dynamic Binary Time Series Models Henri Nyberg University of Helsinki Discussion

More information

Economics Letters 108 (2010) Contents lists available at ScienceDirect. Economics Letters. journal homepage:

Economics Letters 108 (2010) Contents lists available at ScienceDirect. Economics Letters. journal homepage: Economics Letters 108 (2010) 167 171 Contents lists available at ScienceDirect Economics Letters journal homepage: www.elsevier.com/locate/ecolet Is there a financial accelerator in US banking? Evidence

More information

9. Logit and Probit Models For Dichotomous Data

9. Logit and Probit Models For Dichotomous Data Sociology 740 John Fox Lecture Notes 9. Logit and Probit Models For Dichotomous Data Copyright 2014 by John Fox Logit and Probit Models for Dichotomous Responses 1 1. Goals: I To show how models similar

More information

Market Timing Does Work: Evidence from the NYSE 1

Market Timing Does Work: Evidence from the NYSE 1 Market Timing Does Work: Evidence from the NYSE 1 Devraj Basu Alexander Stremme Warwick Business School, University of Warwick November 2005 address for correspondence: Alexander Stremme Warwick Business

More information

ANALYZING MACROECONOMIC FORECASTABILITY. Ray C. Fair. June 2009 Updated: September 2009 COWLES FOUNDATION DISCUSSION PAPER NO.

ANALYZING MACROECONOMIC FORECASTABILITY. Ray C. Fair. June 2009 Updated: September 2009 COWLES FOUNDATION DISCUSSION PAPER NO. ANALYZING MACROECONOMIC FORECASTABILITY By Ray C. Fair June 2009 Updated: September 2009 COWLES FOUNDATION DISCUSSION PAPER NO. 1706 COWLES FOUNDATION FOR RESEARCH IN ECONOMICS YALE UNIVERSITY Box 208281

More information

On the economic significance of stock return predictability: Evidence from macroeconomic state variables

On the economic significance of stock return predictability: Evidence from macroeconomic state variables On the economic significance of stock return predictability: Evidence from macroeconomic state variables Huacheng Zhang * University of Arizona This draft: 8/31/2012 First draft: 2/28/2012 Abstract We

More information

High-Frequency Data Analysis and Market Microstructure [Tsay (2005), chapter 5]

High-Frequency Data Analysis and Market Microstructure [Tsay (2005), chapter 5] 1 High-Frequency Data Analysis and Market Microstructure [Tsay (2005), chapter 5] High-frequency data have some unique characteristics that do not appear in lower frequencies. At this class we have: Nonsynchronous

More information

Forecasting turning points of the business cycle: dynamic logit models for panel data

Forecasting turning points of the business cycle: dynamic logit models for panel data The 9th Biennial Conference of the Czech Economic Society Forecasting turning points of the business cycle: dynamic logit models for panel data Anna Pestova Senior expert, CMASF Research fellow, National

More information

2. Copula Methods Background

2. Copula Methods Background 1. Introduction Stock futures markets provide a channel for stock holders potentially transfer risks. Effectiveness of such a hedging strategy relies heavily on the accuracy of hedge ratio estimation.

More information

Implied Volatility v/s Realized Volatility: A Forecasting Dimension

Implied Volatility v/s Realized Volatility: A Forecasting Dimension 4 Implied Volatility v/s Realized Volatility: A Forecasting Dimension 4.1 Introduction Modelling and predicting financial market volatility has played an important role for market participants as it enables

More information

Investing through Economic Cycles with Ensemble Machine Learning Algorithms

Investing through Economic Cycles with Ensemble Machine Learning Algorithms Investing through Economic Cycles with Ensemble Machine Learning Algorithms Thomas Raffinot Silex Investment Partners Big Data in Finance Conference Thomas Raffinot (Silex-IP) Economic Cycles-Machine Learning

More information

FE670 Algorithmic Trading Strategies. Stevens Institute of Technology

FE670 Algorithmic Trading Strategies. Stevens Institute of Technology FE670 Algorithmic Trading Strategies Lecture 4. Cross-Sectional Models and Trading Strategies Steve Yang Stevens Institute of Technology 09/26/2013 Outline 1 Cross-Sectional Methods for Evaluation of Factor

More information

Amath 546/Econ 589 Univariate GARCH Models: Advanced Topics

Amath 546/Econ 589 Univariate GARCH Models: Advanced Topics Amath 546/Econ 589 Univariate GARCH Models: Advanced Topics Eric Zivot April 29, 2013 Lecture Outline The Leverage Effect Asymmetric GARCH Models Forecasts from Asymmetric GARCH Models GARCH Models with

More information

Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016)

Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 An Investigation of the Structural Characteristics of the Indian IT Sector and the Capital Goods Sector An Application of the

More information

Sharpe Ratio over investment Horizon

Sharpe Ratio over investment Horizon Sharpe Ratio over investment Horizon Ziemowit Bednarek, Pratish Patel and Cyrus Ramezani December 8, 2014 ABSTRACT Both building blocks of the Sharpe ratio the expected return and the expected volatility

More information

Leading indicators of the business cycle: dynamic probit models for OECD countries and Russia

Leading indicators of the business cycle: dynamic probit models for OECD countries and Russia 18th International Conference on Macroeconomic Analysis and International Finance Leading indicators of the business cycle: dynamic probit models for OECD countries and Russia Anna Pestova Research fellow,

More information

Risk-Adjusted Futures and Intermeeting Moves

Risk-Adjusted Futures and Intermeeting Moves issn 1936-5330 Risk-Adjusted Futures and Intermeeting Moves Brent Bundick Federal Reserve Bank of Kansas City First Version: October 2007 This Version: June 2008 RWP 07-08 Abstract Piazzesi and Swanson

More information

Analyzing the Determinants of Project Success: A Probit Regression Approach

Analyzing the Determinants of Project Success: A Probit Regression Approach 2016 Annual Evaluation Review, Linked Document D 1 Analyzing the Determinants of Project Success: A Probit Regression Approach 1. This regression analysis aims to ascertain the factors that determine development

More information

The Comovements Along the Term Structure of Oil Forwards in Periods of High and Low Volatility: How Tight Are They?

The Comovements Along the Term Structure of Oil Forwards in Periods of High and Low Volatility: How Tight Are They? The Comovements Along the Term Structure of Oil Forwards in Periods of High and Low Volatility: How Tight Are They? Massimiliano Marzo and Paolo Zagaglia This version: January 6, 29 Preliminary: comments

More information

Online Appendix to Bond Return Predictability: Economic Value and Links to the Macroeconomy. Pairwise Tests of Equality of Forecasting Performance

Online Appendix to Bond Return Predictability: Economic Value and Links to the Macroeconomy. Pairwise Tests of Equality of Forecasting Performance Online Appendix to Bond Return Predictability: Economic Value and Links to the Macroeconomy This online appendix is divided into four sections. In section A we perform pairwise tests aiming at disentangling

More information

starting on 5/1/1953 up until 2/1/2017.

starting on 5/1/1953 up until 2/1/2017. An Actuary s Guide to Financial Applications: Examples with EViews By William Bourgeois An actuary is a business professional who uses statistics to determine and analyze risks for companies. In this guide,

More information

Premium Timing with Valuation Ratios

Premium Timing with Valuation Ratios RESEARCH Premium Timing with Valuation Ratios March 2016 Wei Dai, PhD Research The predictability of expected stock returns is an old topic and an important one. While investors may increase expected returns

More information

Lecture 8: Markov and Regime

Lecture 8: Markov and Regime Lecture 8: Markov and Regime Switching Models Prof. Massimo Guidolin 20192 Financial Econometrics Spring 2016 Overview Motivation Deterministic vs. Endogeneous, Stochastic Switching Dummy Regressiom Switching

More information

HOUSEHOLDS INDEBTEDNESS: A MICROECONOMIC ANALYSIS BASED ON THE RESULTS OF THE HOUSEHOLDS FINANCIAL AND CONSUMPTION SURVEY*

HOUSEHOLDS INDEBTEDNESS: A MICROECONOMIC ANALYSIS BASED ON THE RESULTS OF THE HOUSEHOLDS FINANCIAL AND CONSUMPTION SURVEY* HOUSEHOLDS INDEBTEDNESS: A MICROECONOMIC ANALYSIS BASED ON THE RESULTS OF THE HOUSEHOLDS FINANCIAL AND CONSUMPTION SURVEY* Sónia Costa** Luísa Farinha** 133 Abstract The analysis of the Portuguese households

More information

A Test of the Normality Assumption in the Ordered Probit Model *

A Test of the Normality Assumption in the Ordered Probit Model * A Test of the Normality Assumption in the Ordered Probit Model * Paul A. Johnson Working Paper No. 34 March 1996 * Assistant Professor, Vassar College. I thank Jahyeong Koo, Jim Ziliak and an anonymous

More information

Five Things You Should Know About Quantile Regression

Five Things You Should Know About Quantile Regression Five Things You Should Know About Quantile Regression Robert N. Rodriguez and Yonggang Yao SAS Institute #analyticsx Copyright 2016, SAS Institute Inc. All rights reserved. Quantile regression brings the

More information

Combining State-Dependent Forecasts of Equity Risk Premium

Combining State-Dependent Forecasts of Equity Risk Premium Combining State-Dependent Forecasts of Equity Risk Premium Daniel de Almeida, Ana-Maria Fuertes and Luiz Koodi Hotta Universidad Carlos III de Madrid September 15, 216 Almeida, Fuertes and Hotta (UC3M)

More information

Fitting financial time series returns distributions: a mixture normality approach

Fitting financial time series returns distributions: a mixture normality approach Fitting financial time series returns distributions: a mixture normality approach Riccardo Bramante and Diego Zappa * Abstract Value at Risk has emerged as a useful tool to risk management. A relevant

More information

Cross-Sectional Distribution of GARCH Coefficients across S&P 500 Constituents : Time-Variation over the Period

Cross-Sectional Distribution of GARCH Coefficients across S&P 500 Constituents : Time-Variation over the Period Cahier de recherche/working Paper 13-13 Cross-Sectional Distribution of GARCH Coefficients across S&P 500 Constituents : Time-Variation over the Period 2000-2012 David Ardia Lennart F. Hoogerheide Mai/May

More information

Mixing Frequencies: Stock Returns as a Predictor of Real Output Growth

Mixing Frequencies: Stock Returns as a Predictor of Real Output Growth SMU ECONOMICS & STATISTICS WORKING PAPER SERIES Mixing Frequencies: Stock Returns as a Predictor of Real Output Growth Anthony S. Tay December 26 Paper No. 34-26 ANY OPINIONS EXPRESSED ARE THOSE OF THE

More information

Internet Appendix for: Cyclical Dispersion in Expected Defaults

Internet Appendix for: Cyclical Dispersion in Expected Defaults Internet Appendix for: Cyclical Dispersion in Expected Defaults March, 2018 Contents 1 1 Robustness Tests The results presented in the main text are robust to the definition of debt repayments, and the

More information

DYNAMIC CORRELATIONS AND FORECASTING OF TERM STRUCTURE SLOPES IN EUROCURRENCY MARKETS

DYNAMIC CORRELATIONS AND FORECASTING OF TERM STRUCTURE SLOPES IN EUROCURRENCY MARKETS DYNAMIC CORRELATIONS AND FORECASTING OF TERM STRUCTURE SLOPES IN EUROCURRENCY MARKETS Emilio Domínguez 1 Alfonso Novales 2 April 1999 ABSTRACT Using monthly data on Euro-rates for 1979-1998, we examine

More information

Lecture 9: Markov and Regime

Lecture 9: Markov and Regime Lecture 9: Markov and Regime Switching Models Prof. Massimo Guidolin 20192 Financial Econometrics Spring 2017 Overview Motivation Deterministic vs. Endogeneous, Stochastic Switching Dummy Regressiom Switching

More information

The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2011, Mr. Ruey S. Tsay. Solutions to Final Exam.

The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2011, Mr. Ruey S. Tsay. Solutions to Final Exam. The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2011, Mr. Ruey S. Tsay Solutions to Final Exam Problem A: (32 pts) Answer briefly the following questions. 1. Suppose

More information

Forecasting Canadian Recessions Using Qual VAR Model. By Lin Chen ( ) Major paper presented to the Department of Economics of the

Forecasting Canadian Recessions Using Qual VAR Model. By Lin Chen ( ) Major paper presented to the Department of Economics of the Forecasting Canadian Recessions Using Qual VAR Model By Lin Chen (7410777) Major paper presented to the Department of Economics of the University of Ottawa in partial fulfillment of the requirements of

More information

The Distributions of Income and Consumption. Risk: Evidence from Norwegian Registry Data

The Distributions of Income and Consumption. Risk: Evidence from Norwegian Registry Data The Distributions of Income and Consumption Risk: Evidence from Norwegian Registry Data Elin Halvorsen Hans A. Holter Serdar Ozkan Kjetil Storesletten February 15, 217 Preliminary Extended Abstract Version

More information

Lecture 1: The Econometrics of Financial Returns

Lecture 1: The Econometrics of Financial Returns Lecture 1: The Econometrics of Financial Returns Prof. Massimo Guidolin 20192 Financial Econometrics Winter/Spring 2016 Overview General goals of the course and definition of risk(s) Predicting asset returns:

More information

INDICATORS OF FINANCIAL DISTRESS IN MATURE ECONOMIES

INDICATORS OF FINANCIAL DISTRESS IN MATURE ECONOMIES B INDICATORS OF FINANCIAL DISTRESS IN MATURE ECONOMIES This special feature analyses the indicator properties of macroeconomic variables and aggregated financial statements from the banking sector in providing

More information

Solving dynamic portfolio choice problems by recursing on optimized portfolio weights or on the value function?

Solving dynamic portfolio choice problems by recursing on optimized portfolio weights or on the value function? DOI 0.007/s064-006-9073-z ORIGINAL PAPER Solving dynamic portfolio choice problems by recursing on optimized portfolio weights or on the value function? Jules H. van Binsbergen Michael W. Brandt Received:

More information

Chapter 7 Selected Answers

Chapter 7 Selected Answers Chapter 7 Selected Answers Problem 7.1: a) When Clorox buy back some of its bonds, fewer bonds are available at each interest rate, so that the borrowing curve in Figure 7.1.1 shifts leftward from Use

More information

Market Variables and Financial Distress. Giovanni Fernandez Stetson University

Market Variables and Financial Distress. Giovanni Fernandez Stetson University Market Variables and Financial Distress Giovanni Fernandez Stetson University In this paper, I investigate the predictive ability of market variables in correctly predicting and distinguishing going concern

More information

Business Statistics 41000: Probability 3

Business Statistics 41000: Probability 3 Business Statistics 41000: Probability 3 Drew D. Creal University of Chicago, Booth School of Business February 7 and 8, 2014 1 Class information Drew D. Creal Email: dcreal@chicagobooth.edu Office: 404

More information

Discussion of Trend Inflation in Advanced Economies

Discussion of Trend Inflation in Advanced Economies Discussion of Trend Inflation in Advanced Economies James Morley University of New South Wales 1. Introduction Garnier, Mertens, and Nelson (this issue, GMN hereafter) conduct model-based trend/cycle decomposition

More information

The complementary nature of ratings and market-based measures of default risk. Gunter Löffler* University of Ulm January 2007

The complementary nature of ratings and market-based measures of default risk. Gunter Löffler* University of Ulm January 2007 The complementary nature of ratings and market-based measures of default risk Gunter Löffler* University of Ulm January 2007 Key words: default prediction, credit ratings, Merton approach. * Gunter Löffler,

More information

The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2017, Mr. Ruey S. Tsay. Solutions to Final Exam

The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2017, Mr. Ruey S. Tsay. Solutions to Final Exam The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2017, Mr. Ruey S. Tsay Solutions to Final Exam Problem A: (40 points) Answer briefly the following questions. 1. Describe

More information

A Statistical Analysis to Predict Financial Distress

A Statistical Analysis to Predict Financial Distress J. Service Science & Management, 010, 3, 309-335 doi:10.436/jssm.010.33038 Published Online September 010 (http://www.scirp.org/journal/jssm) 309 Nicolas Emanuel Monti, Roberto Mariano Garcia Department

More information

Chapter IV. Forecasting Daily and Weekly Stock Returns

Chapter IV. Forecasting Daily and Weekly Stock Returns Forecasting Daily and Weekly Stock Returns An unsophisticated forecaster uses statistics as a drunken man uses lamp-posts -for support rather than for illumination.0 Introduction In the previous chapter,

More information

Online Appendix (Not intended for Publication): Federal Reserve Credibility and the Term Structure of Interest Rates

Online Appendix (Not intended for Publication): Federal Reserve Credibility and the Term Structure of Interest Rates Online Appendix Not intended for Publication): Federal Reserve Credibility and the Term Structure of Interest Rates Aeimit Lakdawala Michigan State University Shu Wu University of Kansas August 2017 1

More information

Equity Price Dynamics Before and After the Introduction of the Euro: A Note*

Equity Price Dynamics Before and After the Introduction of the Euro: A Note* Equity Price Dynamics Before and After the Introduction of the Euro: A Note* Yin-Wong Cheung University of California, U.S.A. Frank Westermann University of Munich, Germany Daily data from the German and

More information

Window Width Selection for L 2 Adjusted Quantile Regression

Window Width Selection for L 2 Adjusted Quantile Regression Window Width Selection for L 2 Adjusted Quantile Regression Yoonsuh Jung, The Ohio State University Steven N. MacEachern, The Ohio State University Yoonkyung Lee, The Ohio State University Technical Report

More information

Forecasting Singapore economic growth with mixed-frequency data

Forecasting Singapore economic growth with mixed-frequency data Edith Cowan University Research Online ECU Publications 2013 2013 Forecasting Singapore economic growth with mixed-frequency data A. Tsui C.Y. Xu Zhaoyong Zhang Edith Cowan University, zhaoyong.zhang@ecu.edu.au

More information

Simple Fuzzy Score for Russian Public Companies Risk of Default

Simple Fuzzy Score for Russian Public Companies Risk of Default Simple Fuzzy Score for Russian Public Companies Risk of Default By Sergey Ivliev April 2,2. Introduction Current economy crisis of 28 29 has resulted in severe credit crunch and significant NPL rise in

More information

The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2009, Mr. Ruey S. Tsay. Solutions to Final Exam

The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2009, Mr. Ruey S. Tsay. Solutions to Final Exam The University of Chicago, Booth School of Business Business 41202, Spring Quarter 2009, Mr. Ruey S. Tsay Solutions to Final Exam Problem A: (42 pts) Answer briefly the following questions. 1. Questions

More information

The Nelson-Siegel-Svensson Model for U.S. Treasury Securities and Its Interpretation

The Nelson-Siegel-Svensson Model for U.S. Treasury Securities and Its Interpretation 1 The Nelson-Siegel-Svensson Model for U.S. Treasury Securities and Its Interpretation By Lisa Patrick 1 Introduction Whether you are an investor in equities, bonds, real estate, or other financial securities,

More information

the display, exploration and transformation of the data are demonstrated and biases typically encountered are highlighted.

the display, exploration and transformation of the data are demonstrated and biases typically encountered are highlighted. 1 Insurance data Generalized linear modeling is a methodology for modeling relationships between variables. It generalizes the classical normal linear model, by relaxing some of its restrictive assumptions,

More information

Do core inflation measures help forecast inflation? Out-of-sample evidence from French data

Do core inflation measures help forecast inflation? Out-of-sample evidence from French data Economics Letters 69 (2000) 261 266 www.elsevier.com/ locate/ econbase Do core inflation measures help forecast inflation? Out-of-sample evidence from French data Herve Le Bihan *, Franck Sedillot Banque

More information

The Enhanced Aggregate Spread -- An Owner s Manual. Robert F. Dieli, Ph.D.

The Enhanced Aggregate Spread -- An Owner s Manual. Robert F. Dieli, Ph.D. The Enhanced Aggregate Spread -- An Owner s Manual Robert F. Dieli, Ph.D. Chart 1 The Phases of the Business Cycle Mr. Model is a forecasting tool that was developed to do two things. First, provide a

More information

Discussion of The Role of Expectations in Inflation Dynamics

Discussion of The Role of Expectations in Inflation Dynamics Discussion of The Role of Expectations in Inflation Dynamics James H. Stock Department of Economics, Harvard University and the NBER 1. Introduction Rational expectations are at the heart of the dynamic

More information

Easy and Successful Macroeconomic Timing

Easy and Successful Macroeconomic Timing Easy and Successful Macroeconomic Timing William Rafter, MathInvest LLC Abstract When the economy takes a turn for the worse, employment declines, right? Well, not all employment. Certainly, full-time

More information

Yield Curve and Predicted GDP Growth, September 2017

Yield Curve and Predicted GDP Growth, September 2017 1 6 Yield Curve and Predicted GDP Growth, September 2017 Latest Data Archives Covering September 23, 2017 October 20, 2017 Highlights October September August 3-month Treasury bill rate (percent) 1.10

More information

ME3620. Theory of Engineering Experimentation. Spring Chapter III. Random Variables and Probability Distributions.

ME3620. Theory of Engineering Experimentation. Spring Chapter III. Random Variables and Probability Distributions. ME3620 Theory of Engineering Experimentation Chapter III. Random Variables and Probability Distributions Chapter III 1 3.2 Random Variables In an experiment, a measurement is usually denoted by a variable

More information

Business Cycles. Trends and cycles. Overview. Trends and cycles. Chris Edmond NYU Stern. Spring Start by looking at quarterly US real GDP

Business Cycles. Trends and cycles. Overview. Trends and cycles. Chris Edmond NYU Stern. Spring Start by looking at quarterly US real GDP Trends and cycles Business Cycles Start by looking at quarterly US real Chris Edmond NYU Stern Spring 2007 1 3 Overview Trends and cycles Business cycle properties does not grow smoothly: booms and recessions

More information

Does Commodity Price Index predict Canadian Inflation?

Does Commodity Price Index predict Canadian Inflation? 2011 年 2 月第十四卷一期 Vol. 14, No. 1, February 2011 Does Commodity Price Index predict Canadian Inflation? Tao Chen http://cmr.ba.ouhk.edu.hk Web Journal of Chinese Management Review Vol. 14 No 1 1 Does Commodity

More information

The mean-variance portfolio choice framework and its generalizations

The mean-variance portfolio choice framework and its generalizations The mean-variance portfolio choice framework and its generalizations Prof. Massimo Guidolin 20135 Theory of Finance, Part I (Sept. October) Fall 2014 Outline and objectives The backward, three-step solution

More information

Modelling catastrophic risk in international equity markets: An extreme value approach. JOHN COTTER University College Dublin

Modelling catastrophic risk in international equity markets: An extreme value approach. JOHN COTTER University College Dublin Modelling catastrophic risk in international equity markets: An extreme value approach JOHN COTTER University College Dublin Abstract: This letter uses the Block Maxima Extreme Value approach to quantify

More information

Forecasting U.S. Recessions and Economic Activity

Forecasting U.S. Recessions and Economic Activity Forecasting U.S. Recessions and Economic Activity Rachidi Kotchoni Dalibor Stevanovic This version: February 9, 28 Abstract This paper proposes a simple nonlinear framework to produce real-time multi-horizon

More information

Int. Statistical Inst.: Proc. 58th World Statistical Congress, 2011, Dublin (Session CPS001) p approach

Int. Statistical Inst.: Proc. 58th World Statistical Congress, 2011, Dublin (Session CPS001) p approach Int. Statistical Inst.: Proc. 58th World Statistical Congress, 2011, Dublin (Session CPS001) p.5901 What drives short rate dynamics? approach A functional gradient descent Audrino, Francesco University

More information

Some Characteristics of Data

Some Characteristics of Data Some Characteristics of Data Not all data is the same, and depending on some characteristics of a particular dataset, there are some limitations as to what can and cannot be done with that data. Some key

More information

I. BACKGROUND AND CONTEXT

I. BACKGROUND AND CONTEXT Review of the Debt Sustainability Framework for Low Income Countries (LIC DSF) Discussion Note August 1, 2016 I. BACKGROUND AND CONTEXT 1. The LIC DSF, introduced in 2005, remains the cornerstone of assessing

More information

Characteristics of the euro area business cycle in the 1990s

Characteristics of the euro area business cycle in the 1990s Characteristics of the euro area business cycle in the 1990s As part of its monetary policy strategy, the ECB regularly monitors the development of a wide range of indicators and assesses their implications

More information

Modeling Private Firm Default: PFirm

Modeling Private Firm Default: PFirm Modeling Private Firm Default: PFirm Grigoris Karakoulas Business Analytic Solutions May 30 th, 2002 Outline Problem Statement Modelling Approaches Private Firm Data Mining Model Development Model Evaluation

More information

Augmenting Okun s Law with Earnings and the Unemployment Puzzle of 2011

Augmenting Okun s Law with Earnings and the Unemployment Puzzle of 2011 Augmenting Okun s Law with Earnings and the Unemployment Puzzle of 2011 Kurt G. Lunsford University of Wisconsin Madison January 2013 Abstract I propose an augmented version of Okun s law that regresses

More information

Key Words: emerging markets, copulas, tail dependence, Value-at-Risk JEL Classification: C51, C52, C14, G17

Key Words: emerging markets, copulas, tail dependence, Value-at-Risk JEL Classification: C51, C52, C14, G17 RISK MANAGEMENT WITH TAIL COPULAS FOR EMERGING MARKET PORTFOLIOS Svetlana Borovkova Vrije Universiteit Amsterdam Faculty of Economics and Business Administration De Boelelaan 1105, 1081 HV Amsterdam, The

More information

Inflation Targeting and Revisions to Inflation Data: A Case Study with PCE Inflation * Calvin Price July 2011

Inflation Targeting and Revisions to Inflation Data: A Case Study with PCE Inflation * Calvin Price July 2011 Inflation Targeting and Revisions to Inflation Data: A Case Study with PCE Inflation * Calvin Price July 2011 Introduction Central banks around the world have come to recognize the importance of maintaining

More information

A Multifrequency Theory of the Interest Rate Term Structure

A Multifrequency Theory of the Interest Rate Term Structure A Multifrequency Theory of the Interest Rate Term Structure Laurent Calvet, Adlai Fisher, and Liuren Wu HEC, UBC, & Baruch College Chicago University February 26, 2010 Liuren Wu (Baruch) Cascade Dynamics

More information

Model Construction & Forecast Based Portfolio Allocation:

Model Construction & Forecast Based Portfolio Allocation: QBUS6830 Financial Time Series and Forecasting Model Construction & Forecast Based Portfolio Allocation: Is Quantitative Method Worth It? Members: Bowei Li (303083) Wenjian Xu (308077237) Xiaoyun Lu (3295347)

More information

Identifying Superior Performing Equity Mutual Funds

Identifying Superior Performing Equity Mutual Funds Identifying Superior Performing Equity Mutual Funds Ravi Shukla Finance Department Syracuse University Syracuse, NY 13244-2130 Phone: (315) 443-3576 Email: rkshukla@som.syr.edu First draft: March 1999

More information

Investment Platforms Market Study Interim Report: Annex 7 Fund Discounts and Promotions

Investment Platforms Market Study Interim Report: Annex 7 Fund Discounts and Promotions MS17/1.2: Annex 7 Market Study Investment Platforms Market Study Interim Report: Annex 7 Fund Discounts and Promotions July 2018 Annex 7: Introduction 1. There are several ways in which investment platforms

More information

Using survival models for profit and loss estimation. Dr Tony Bellotti Lecturer in Statistics Department of Mathematics Imperial College London

Using survival models for profit and loss estimation. Dr Tony Bellotti Lecturer in Statistics Department of Mathematics Imperial College London Using survival models for profit and loss estimation Dr Tony Bellotti Lecturer in Statistics Department of Mathematics Imperial College London Credit Scoring and Credit Control XIII conference August 28-30,

More information

Contrarian Trades and Disposition Effect: Evidence from Online Trade Data. Abstract

Contrarian Trades and Disposition Effect: Evidence from Online Trade Data. Abstract Contrarian Trades and Disposition Effect: Evidence from Online Trade Data Hayato Komai a Ryota Koyano b Daisuke Miyakawa c Abstract Using online stock trading records in Japan for 461 individual investors

More information

Intro to GLM Day 2: GLM and Maximum Likelihood

Intro to GLM Day 2: GLM and Maximum Likelihood Intro to GLM Day 2: GLM and Maximum Likelihood Federico Vegetti Central European University ECPR Summer School in Methods and Techniques 1 / 32 Generalized Linear Modeling 3 steps of GLM 1. Specify the

More information

Rating Efficiency in the Indian Commercial Paper Market. Anand Srinivasan 1

Rating Efficiency in the Indian Commercial Paper Market. Anand Srinivasan 1 Rating Efficiency in the Indian Commercial Paper Market Anand Srinivasan 1 Abstract: This memo examines the efficiency of the rating system for commercial paper (CP) issues in India, for issues rated A1+

More information

Forecasting Macedonian Business Cycle Turning Points Using Qual Var Model 1

Forecasting Macedonian Business Cycle Turning Points Using Qual Var Model 1 Forecasting Macedonian Business Cycle Turning Points Using Qual Var Model 61 UDK: 330.1:65.012.511(497.7) DOI: 10.1515/jcbtp-2016-0020 Journal of Central Banking Theory and Practice, 2016, 3, pp. 61-78

More information

Bloomberg. Portfolio Value-at-Risk. Sridhar Gollamudi & Bryan Weber. September 22, Version 1.0

Bloomberg. Portfolio Value-at-Risk. Sridhar Gollamudi & Bryan Weber. September 22, Version 1.0 Portfolio Value-at-Risk Sridhar Gollamudi & Bryan Weber September 22, 2011 Version 1.0 Table of Contents 1 Portfolio Value-at-Risk 2 2 Fundamental Factor Models 3 3 Valuation methodology 5 3.1 Linear factor

More information

Two New Indexes Offer a Broad View of Economic Activity in the New York New Jersey Region

Two New Indexes Offer a Broad View of Economic Activity in the New York New Jersey Region C URRENT IN ECONOMICS FEDERAL RESERVE BANK OF NEW YORK Second I SSUES AND FINANCE district highlights Volume 5 Number 14 October 1999 Two New Indexes Offer a Broad View of Economic Activity in the New

More information

AP STATISTICS FALL SEMESTSER FINAL EXAM STUDY GUIDE

AP STATISTICS FALL SEMESTSER FINAL EXAM STUDY GUIDE AP STATISTICS Name: FALL SEMESTSER FINAL EXAM STUDY GUIDE Period: *Go over Vocabulary Notecards! *This is not a comprehensive review you still should look over your past notes, homework/practice, Quizzes,

More information

MOODY S KMV RISKCALC V3.2 JAPAN

MOODY S KMV RISKCALC V3.2 JAPAN MCH 25, 2009 MOODY S KMV RISKCALC V3.2 JAPAN MODELINGMETHODOLOGY ABSTRACT AUTHORS Lee Chua Douglas W. Dwyer Andrew Zhang Moody s KMV RiskCalc is the Moody's KMV model for predicting private company defaults..

More information

Inflation Regimes and Monetary Policy Surprises in the EU

Inflation Regimes and Monetary Policy Surprises in the EU Inflation Regimes and Monetary Policy Surprises in the EU Tatjana Dahlhaus Danilo Leiva-Leon November 7, VERY PRELIMINARY AND INCOMPLETE Abstract This paper assesses the effect of monetary policy during

More information

The Persistent Effect of Temporary Affirmative Action: Online Appendix

The Persistent Effect of Temporary Affirmative Action: Online Appendix The Persistent Effect of Temporary Affirmative Action: Online Appendix Conrad Miller Contents A Extensions and Robustness Checks 2 A. Heterogeneity by Employer Size.............................. 2 A.2

More information

Financial Econometrics Notes. Kevin Sheppard University of Oxford

Financial Econometrics Notes. Kevin Sheppard University of Oxford Financial Econometrics Notes Kevin Sheppard University of Oxford Monday 15 th January, 2018 2 This version: 22:52, Monday 15 th January, 2018 2018 Kevin Sheppard ii Contents 1 Probability, Random Variables

More information

The use of real-time data is critical, for the Federal Reserve

The use of real-time data is critical, for the Federal Reserve Capacity Utilization As a Real-Time Predictor of Manufacturing Output Evan F. Koenig Research Officer Federal Reserve Bank of Dallas The use of real-time data is critical, for the Federal Reserve indices

More information

Internet Appendix for: Cyclical Dispersion in Expected Defaults

Internet Appendix for: Cyclical Dispersion in Expected Defaults Internet Appendix for: Cyclical Dispersion in Expected Defaults João F. Gomes Marco Grotteria Jessica Wachter August, 2017 Contents 1 Robustness Tests 2 1.1 Multivariable Forecasting of Macroeconomic Quantities............

More information

Historical Trends in the Degree of Federal Income Tax Progressivity in the United States

Historical Trends in the Degree of Federal Income Tax Progressivity in the United States Kennesaw State University DigitalCommons@Kennesaw State University Faculty Publications 5-14-2012 Historical Trends in the Degree of Federal Income Tax Progressivity in the United States Timothy Mathews

More information

Crowdfunding, Cascades and Informed Investors

Crowdfunding, Cascades and Informed Investors DISCUSSION PAPER SERIES IZA DP No. 7994 Crowdfunding, Cascades and Informed Investors Simon C. Parker February 2014 Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor Crowdfunding,

More information

Leading Economic Indicators and a Probabilistic Approach to Estimating Market Tail Risk

Leading Economic Indicators and a Probabilistic Approach to Estimating Market Tail Risk Leading Economic Indicators and a Probabilistic Approach to Estimating Market Tail Risk Sonu Vanrghese, Ph.D. Director of Research Angshuman Gooptu Senior Economist The shifting trends observed in leading

More information

Subject CS1 Actuarial Statistics 1 Core Principles. Syllabus. for the 2019 exams. 1 June 2018

Subject CS1 Actuarial Statistics 1 Core Principles. Syllabus. for the 2019 exams. 1 June 2018 ` Subject CS1 Actuarial Statistics 1 Core Principles Syllabus for the 2019 exams 1 June 2018 Copyright in this Core Reading is the property of the Institute and Faculty of Actuaries who are the sole distributors.

More information